Why the “Best Exercise” Ranking Is Not a Finish Line
The network meta-analysis addresses a different problem from the dose curve: not how much structured activity is associated with an outcome, but how several exercise-based approaches compare for postpartum depressive symptoms. This chapter examines what that network can legitimately rank, why structured aerobic exercise has clearer statistically supported comparisons than its position alone suggests, and why mind–body exercise can rank first without being proven the universal winner. The useful judgment is not “choose the top row,” but “which comparison, population, and certainty make an option defensible here?”
What the network is actually comparing
A conventional pairwise meta-analysis might combine trials of structured aerobic exercise against usual care. A network meta-analysis expands the question by connecting several interventions through both direct and indirect comparisons. If some trials compare aerobic exercise with a control and others compare mind–body exercise with the same kind of control, the network can estimate a relative relationship between the two modalities even when they were not tested head-to-head in every study. That extra reach is useful, but it depends on an assumption called transitivity: the studies must be similar enough in the factors that could change the result for the indirect comparison to be meaningful.
Consider the logic in concrete terms. One group of trials may estimate how SAE differs from passive control, while another estimates how MBE differs from passive control. The network can use the shared control as a bridge. But the bridge is trustworthy only if the participants, timing after childbirth, baseline symptom burden, outcome measurement, program intensity, and other effect-changing features are sufficiently comparable. If the SAE trials mostly involve prevention and the MBE trials mostly involve women already depressed, the shared control does not magically make the populations equivalent. The numerical comparison may still be computable, but its clinical meaning becomes less secure. Indirect evidence expands the map; it does not remove the terrain between the studies.
This is why the number of nodes should not be confused with the number of equally strong comparisons. A network may place many interventions in one diagram while some links are supported by direct head-to-head trials and others mainly by inference. The network’s overall consistency test can ask whether direct and indirect estimates conflict statistically, but an absence of detected inconsistency is not proof that every clinical difference has been accounted for. The reader still needs to ask which route produced the comparison and how much evidence sits behind it.
The postpartum network included 23 randomized controlled trials involving 2,752 women recruited within the first 12 months after childbirth. It organized the evidence into eight nodes. Structured aerobic exercise, or SAE, represented a prespecified aerobic program. Mind–body exercise, or MBE, included approaches such as yoga, breathing regulation, mindfulness, and related psychological-regulation components. The other nodes covered lifestyle physical-activity promotion, resistance training, pelvic-floor muscle training, exercise combined with other interventions, and two forms of control. There were ten direct comparisons connecting these categories.
These labels are analytical tools, not interchangeable brands. SAE and MBE may both involve movement, but MBE also includes elements intended to regulate attention, breathing, emotion, or stress. Exercise combined with other interventions—such as diet, psychological support, health promotion, or behavioral management—is a package.